Summer 2025 - Spring 2026
Excellent Student Award
FPT University
Recognized for 3 consecutive semesters of strong academic performance.
Open to Opportunities 2026
Da Nang, Vietnam · FPT University
Đặng Quang Nhật
AI Undergraduate
About Me
Education & Leadership
AI Major· FPT University Da Nang
2024 — Present
GPA 8.93 / 10.0 · Computer Vision & ML Systems
President· DEVER Developer Club
2025 — Present
Leading 100+ developers building AI solutions
Technical Workshops· DEVER Club
2025 — 2026
Organized AI-assisted software development and community technical sessions
0.00
GPA / 10.0
0+
Research Projects
0
Technical Workshops
100+
DEVER Members
Beyond Projects
Activities & Leadership
Dec 2025 — Present
President
FU-DEVER Technology Club · FPT University Da Nang
Lead a 100+ member student technology club across executive operations, member development, and technical activities.
100+ MembersAI WorkshopsITSCGDG Cloud
Jan 2026 — Apr 2026
Project Lead
Shine Team — "Bật nụ cười lên" Charity Project · Da Nang, Vietnam
Led a 14-member team responsible for planning, budgeting, fundraising, and school coordination; delivered 120+ gifts for approximately 60 attendees.
14 Members120+ GiftsVND 12M+ Raised
May 2025 — Jul 2025
Volunteer Trainer & Mentor
FPT University Da Nang Robotics Team · Da Nang, Vietnam
Mentored students in AWS PartyRock, LLMs, AI agents, image generation, Arduino assembly, and AI-assisted coding for FPTU AI & ROBOTICS CHALLENGE 2025.
National FinalsAWS PartyRockArduinoAI Agents
Recognition
Honors & Awards
2024
Second Prize
Da Nang City-level Mathematics Competition
Awarded in the city-level competition for excellent students.
2024
Third Prize
School-level English Debate Contest
Recognized for argumentation, presentation, and English communication.
Expertise
What I Do
Bridging rigorous research with practical engineering across these four domains.
Computer Vision & Detection
Novel metrics and architectures for tiny object detection, super-resolution pipelines, and scale-invariant evaluation — PyTorch, torchvision, COCO tooling.
Scientific Retrieval & RAG
Evidence-grounded search systems with BM25, BGE dense retrieval, Adaptive RRF, SGAF benchmarking, citation-aware answer generation, and notebook-style document workspaces.
Deep Learning Research
Full research lifecycle from problem formulation through controlled ablation to reproducible benchmarking — attention mechanisms, multi-backbone architectures, mixed-precision training.
Full-Stack AI Applications
End-to-end systems shipping research to production — FastAPI backends, React dashboards, Docker containers, GCP deployment, real-time data pipelines.
Toolbelt
Technical Skills & Stack
ML / DL Frameworks
PyTorch
torchvision
Hugging Face
scikit-learn
LightGBM
XGBoost
Computer Vision
YOLO11
Faster R-CNN
OpenCV
Object Tracking
COCO eval
Retrieval & LLM Systems
BM25
Dense Retrieval
Adaptive RRF
BGE-base
Grounded Citations
Data & Visualization
NumPy
Pandas
Matplotlib
Seaborn
Systems & Cloud
Git
Docker
Vercel
GCP
FastAPI
FFmpeg
Databases
MySQL
MongoDB
Languages
Python
SQL
Featured Projects
Applied AIAI / ML Systems Engineer
TrafficFlow — AI-Powered Vehicle Counting Platform
May 2026 – Jul 2026
End-to-end traffic analytics platform for uploaded videos and live YouTube/HLS streams, with ROI preprocessing, YOLO GPU inference, tracking, lane association, counting, and annotated output rendering.
YOLO11FastAPIDockerFFmpegByteTrackCeleryRedis
- Built lane-level detection, tracking, ROI lane association, and direction-aware line-crossing counting
- Improved Event F1 from 0.835 to 0.942 while reducing ID switches from 169 to 42
- Sustained 14.9 FPS over a 30-minute 1080p HLS stream with 0% dropped or stale frames
Applied AIRetrieval / Full-Stack Engineer
SEG Search Engine — Scientific Retrieval & Document RAG
May 2026 – Jul 2026
Serverless scientific search and notebook RAG application for web search, uploaded corpora, grounded answers, and retrieval-mode comparison, backed by the SEG benchmark work on BGE, BM25, Adaptive RRF, and SGAF.
Next.jsSupabaseBM25BGE-baseAdaptive RRFGroqVercel
- Shipped a Vercel web app with separate web-search sessions and notebook dataset workspaces
- Implemented BM25 and Adaptive RRF retrieval with BGE-base embeddings for evidence-grounded answers
- Benchmarked SEG retrieval on SciFact, NFCorpus, FiQA, and SciDocs, with Frozen P3 SGAF reaching 0.4524 average nDCG@10
ResearchResearch Lead
Tiny Object Detection — ALW and SA-ALW Metrics
Jan 2026 – Present
Bounding-box similarity metrics for maritime tiny-object detection on SOD-TinyPeopleInSea, where 92% of objects are smaller than 32 px.
PyTorchFaster R-CNNResNet-50-FPNRFLACOCO Eval
- Designed ALW with anisotropic position normalization and log-ratio shape terms
- Extended ALW to SA-ALW with parameter-free scale-adaptive temperature and position weighting
- SA-ALW full reached test AP 0.0978 and AP50 0.3059, improving over IGWD by +11.9% AP and +15.8% AP50
ResearchMultimodal Research Engineer
Audio-Visual Contact Classification for Tree Structures
Jul 2026 – Present
Sealed multimodal fusion pipeline for 4-class contact material classification in agricultural tree structures, combining frozen audio segment predictions, wav2vec2 audio features, and CLIP image embeddings on a held-out robot domain.
PyTorchwav2vec2CLIPscikit-learnMultimodal FusionAblation
- Built a four-step cascade over frozen audio predictions, contact detectors, secondary CLIP trunk rescue, and leaf/twig material redecode
- Locked hyperparameters on hand data only, then opened the robot/test split once for a sealed claim
- Reached 0.850781 Macro F1 on 2,219 robot/test windows with 0.891393 4-class accuracy
ResearchResearch Software Contributor
MLBII — Continuous-Gap Time-Series Imputation
Jul 2026 – Present
Reproducibility repository for a group paper on continuous-gap multivariate time-series imputation, combining bidirectional temporal lag features, cross-variable context, iterative refinement, and weighted directional fusion.
PythonLightGBMscikit-learnPyPOTSBRITSSAITS
- Co-developed a reproducible benchmark suite for MLBII-Ridge, MLBII-LightGBM, MICE, MissForest, BRITS, and SAITS
- Covered public meteorological and ETT benchmark datasets with fixed-seed continuous gaps of 1, 3, 5, and 7 days
- Packaged command-line experiments plus standalone ETTm1 notebooks for Colab, Kaggle, JupyterLab, and GitHub review
Research Manuscripts
Research & Working Papers
Research CodeWorking PaperWorking Paper
ALW and SA-ALW: Scale-Adaptive Gaussian Metrics for Tiny Object Detection
Dang Quang Nhat · 2026
0.0978 AP
Key Result
SA-ALW full reaches test AP 0.0978 and AP50 0.3059, improving over IGWD
Bounding-box similarity metrics for maritime tiny-object detection on SOD-TinyPeopleInSea. ALW combines anisotropic position normalization with log-ratio shape terms; SA-ALW adds parameter-free scale adaptation for objects from 2 px to 288 px and is evaluated under COCO-style locked-test metrics.
Computer VisionTiny Object DetectionGaussian Metrics
Sealed ClaimWorking Paper
Audio-Visual Fusion for Contact Material Classification in Tree Structures
Dang Quang Nhat · 2026
0.851 F1
Key Result
Macro F1 0.850781 and 4-class accuracy 0.891393 on 2,219 robot/test windows
Multimodal fusion pipeline for robot-domain contact material classification across ambient, leaf, trunk, and twig windows. The claim path uses hand-only hyperparameter selection, frozen audio segment predictions, wav2vec2 audio features, CLIP image embeddings, and a one-shot sealed robot/test evaluation.
Multimodal LearningAudio-Visual FusionAgricultural Robotics
Group PaperWorking Paper
MLBII: Bidirectional Temporal Feature Fusion for Continuous-Gap Multivariate Time-Series Imputation
Le Ho Anh Duy, Dang Quang Nhat, Pham Minh Tien, Hien Nguyen Thi, Thi-Thu-Hong Phan · 2026
6 x 7
Key Result
Reproducible protocol across 6 public datasets, 7 methods, and 1/3/5/7-day continuous gaps
MLBII reconstructs continuous missing intervals by combining forward and backward lag features, contemporaneous cross-variable information, iterative refinement, and context-weighted directional fusion. The repository includes public meteorological and ETT benchmarks with comparable MICE, MissForest, BRITS, and SAITS baselines.
Time-Series ImputationBenchmarkingResearch Software
Da Nang, Vietnam
Open to Internships & Collaborations
Let's Connect
Interested in collaborating on research, building AI systems, or just want to say hello? I'd love to hear from you.
Send Emailor find me on